I am interested to hear thoughts regarding back-testing of trading ideas and "how much is enough".
Suppose one has a trading idea and uses historical trading data to back-test it, how many trades or weeks of history are sufficient for the data/conclusions about the system to be meaningful? I suspect that this is "the million dollar question" but would be happy to receive your insights.
For example, if I have devised a system that is profitable over 3000 trades across 4 currency pairs and has a "winrate" of 75%, is it fair to say that the backtesting on those 3000 trades will produce meaningful/reliable results because the winrate is quite high?
I am also worried about the danger of "tuning" a strategy to the back-test data to the point where it falls apart when exposed to "fresh" market data that the model hasn't seen before. Is it also fair to say that if one has more than a certain number of trades, the risk of this occurring diminishes somewhat?